On Thu, May 01, 2025 at 06:00:12AM +0100, John Garry wrote:

On 01/05/2025 05:30, Darrick J. Wong wrote:

quoted

On Wed, Apr 30, 2025 at 08:52:00AM +0100, John Garry wrote:

quoted

On 25/04/2025 17:45, John Garry wrote:

quoted

+static inline xfs_extlen_t xfs_calc_perag_awu_max(struct xfs_mount *mp)
+{
+	if (mp->m_ddev_targp->bt_bdev_awu_min > 0)
+		return max_pow_of_two_factor(mp->m_sb.sb_agblocks);
+	return mp->m_ag_max_usable;

I think that this should be rounddown_pow_of_two(mp->m_ag_max_usable)

ditto for rt

I will fix (unless disagree).

I don't think this needs fixing.  If there's no hardware support on the
device, then we can do any size of atomic write that we want.

Check man pages for statx:

stx_atomic_write_unit_min
stx_atomic_write_unit_max
              ... These values are each guaranteed to be
              a power-of-2.

Same is enforced for size for RWF_ATOMIC.

Ok then.

--D